]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * IPVS: Destination Hashing scheduling module | |
3 | * | |
4 | * Version: $Id: ip_vs_dh.c,v 1.5 2002/09/15 08:14:08 wensong Exp $ | |
5 | * | |
6 | * Authors: Wensong Zhang <wensong@gnuchina.org> | |
7 | * | |
8 | * Inspired by the consistent hashing scheduler patch from | |
9 | * Thomas Proell <proellt@gmx.de> | |
10 | * | |
11 | * This program is free software; you can redistribute it and/or | |
12 | * modify it under the terms of the GNU General Public License | |
13 | * as published by the Free Software Foundation; either version | |
14 | * 2 of the License, or (at your option) any later version. | |
15 | * | |
16 | * Changes: | |
17 | * | |
18 | */ | |
19 | ||
20 | /* | |
21 | * The dh algorithm is to select server by the hash key of destination IP | |
22 | * address. The pseudo code is as follows: | |
23 | * | |
24 | * n <- servernode[dest_ip]; | |
25 | * if (n is dead) OR | |
26 | * (n is overloaded) OR (n.weight <= 0) then | |
27 | * return NULL; | |
28 | * | |
29 | * return n; | |
30 | * | |
31 | * Notes that servernode is a 256-bucket hash table that maps the hash | |
32 | * index derived from packet destination IP address to the current server | |
33 | * array. If the dh scheduler is used in cache cluster, it is good to | |
34 | * combine it with cache_bypass feature. When the statically assigned | |
35 | * server is dead or overloaded, the load balancer can bypass the cache | |
36 | * server and send requests to the original server directly. | |
37 | * | |
38 | */ | |
39 | ||
40 | #include <linux/module.h> | |
41 | #include <linux/kernel.h> | |
42 | ||
43 | #include <net/ip_vs.h> | |
44 | ||
45 | ||
46 | /* | |
47 | * IPVS DH bucket | |
48 | */ | |
49 | struct ip_vs_dh_bucket { | |
50 | struct ip_vs_dest *dest; /* real server (cache) */ | |
51 | }; | |
52 | ||
53 | /* | |
54 | * for IPVS DH entry hash table | |
55 | */ | |
56 | #ifndef CONFIG_IP_VS_DH_TAB_BITS | |
57 | #define CONFIG_IP_VS_DH_TAB_BITS 8 | |
58 | #endif | |
59 | #define IP_VS_DH_TAB_BITS CONFIG_IP_VS_DH_TAB_BITS | |
60 | #define IP_VS_DH_TAB_SIZE (1 << IP_VS_DH_TAB_BITS) | |
61 | #define IP_VS_DH_TAB_MASK (IP_VS_DH_TAB_SIZE - 1) | |
62 | ||
63 | ||
64 | /* | |
65 | * Returns hash value for IPVS DH entry | |
66 | */ | |
67 | static inline unsigned ip_vs_dh_hashkey(__u32 addr) | |
68 | { | |
69 | return (ntohl(addr)*2654435761UL) & IP_VS_DH_TAB_MASK; | |
70 | } | |
71 | ||
72 | ||
73 | /* | |
74 | * Get ip_vs_dest associated with supplied parameters. | |
75 | */ | |
76 | static inline struct ip_vs_dest * | |
77 | ip_vs_dh_get(struct ip_vs_dh_bucket *tbl, __u32 addr) | |
78 | { | |
79 | return (tbl[ip_vs_dh_hashkey(addr)]).dest; | |
80 | } | |
81 | ||
82 | ||
83 | /* | |
84 | * Assign all the hash buckets of the specified table with the service. | |
85 | */ | |
86 | static int | |
87 | ip_vs_dh_assign(struct ip_vs_dh_bucket *tbl, struct ip_vs_service *svc) | |
88 | { | |
89 | int i; | |
90 | struct ip_vs_dh_bucket *b; | |
91 | struct list_head *p; | |
92 | struct ip_vs_dest *dest; | |
93 | ||
94 | b = tbl; | |
95 | p = &svc->destinations; | |
96 | for (i=0; i<IP_VS_DH_TAB_SIZE; i++) { | |
97 | if (list_empty(p)) { | |
98 | b->dest = NULL; | |
99 | } else { | |
100 | if (p == &svc->destinations) | |
101 | p = p->next; | |
102 | ||
103 | dest = list_entry(p, struct ip_vs_dest, n_list); | |
104 | atomic_inc(&dest->refcnt); | |
105 | b->dest = dest; | |
106 | ||
107 | p = p->next; | |
108 | } | |
109 | b++; | |
110 | } | |
111 | return 0; | |
112 | } | |
113 | ||
114 | ||
115 | /* | |
116 | * Flush all the hash buckets of the specified table. | |
117 | */ | |
118 | static void ip_vs_dh_flush(struct ip_vs_dh_bucket *tbl) | |
119 | { | |
120 | int i; | |
121 | struct ip_vs_dh_bucket *b; | |
122 | ||
123 | b = tbl; | |
124 | for (i=0; i<IP_VS_DH_TAB_SIZE; i++) { | |
125 | if (b->dest) { | |
126 | atomic_dec(&b->dest->refcnt); | |
127 | b->dest = NULL; | |
128 | } | |
129 | b++; | |
130 | } | |
131 | } | |
132 | ||
133 | ||
134 | static int ip_vs_dh_init_svc(struct ip_vs_service *svc) | |
135 | { | |
136 | struct ip_vs_dh_bucket *tbl; | |
137 | ||
138 | /* allocate the DH table for this service */ | |
139 | tbl = kmalloc(sizeof(struct ip_vs_dh_bucket)*IP_VS_DH_TAB_SIZE, | |
140 | GFP_ATOMIC); | |
141 | if (tbl == NULL) { | |
142 | IP_VS_ERR("ip_vs_dh_init_svc(): no memory\n"); | |
143 | return -ENOMEM; | |
144 | } | |
145 | svc->sched_data = tbl; | |
146 | IP_VS_DBG(6, "DH hash table (memory=%Zdbytes) allocated for " | |
147 | "current service\n", | |
148 | sizeof(struct ip_vs_dh_bucket)*IP_VS_DH_TAB_SIZE); | |
149 | ||
150 | /* assign the hash buckets with the updated service */ | |
151 | ip_vs_dh_assign(tbl, svc); | |
152 | ||
153 | return 0; | |
154 | } | |
155 | ||
156 | ||
157 | static int ip_vs_dh_done_svc(struct ip_vs_service *svc) | |
158 | { | |
159 | struct ip_vs_dh_bucket *tbl = svc->sched_data; | |
160 | ||
161 | /* got to clean up hash buckets here */ | |
162 | ip_vs_dh_flush(tbl); | |
163 | ||
164 | /* release the table itself */ | |
165 | kfree(svc->sched_data); | |
166 | IP_VS_DBG(6, "DH hash table (memory=%Zdbytes) released\n", | |
167 | sizeof(struct ip_vs_dh_bucket)*IP_VS_DH_TAB_SIZE); | |
168 | ||
169 | return 0; | |
170 | } | |
171 | ||
172 | ||
173 | static int ip_vs_dh_update_svc(struct ip_vs_service *svc) | |
174 | { | |
175 | struct ip_vs_dh_bucket *tbl = svc->sched_data; | |
176 | ||
177 | /* got to clean up hash buckets here */ | |
178 | ip_vs_dh_flush(tbl); | |
179 | ||
180 | /* assign the hash buckets with the updated service */ | |
181 | ip_vs_dh_assign(tbl, svc); | |
182 | ||
183 | return 0; | |
184 | } | |
185 | ||
186 | ||
187 | /* | |
188 | * If the dest flags is set with IP_VS_DEST_F_OVERLOAD, | |
189 | * consider that the server is overloaded here. | |
190 | */ | |
191 | static inline int is_overloaded(struct ip_vs_dest *dest) | |
192 | { | |
193 | return dest->flags & IP_VS_DEST_F_OVERLOAD; | |
194 | } | |
195 | ||
196 | ||
197 | /* | |
198 | * Destination hashing scheduling | |
199 | */ | |
200 | static struct ip_vs_dest * | |
201 | ip_vs_dh_schedule(struct ip_vs_service *svc, const struct sk_buff *skb) | |
202 | { | |
203 | struct ip_vs_dest *dest; | |
204 | struct ip_vs_dh_bucket *tbl; | |
205 | struct iphdr *iph = skb->nh.iph; | |
206 | ||
207 | IP_VS_DBG(6, "ip_vs_dh_schedule(): Scheduling...\n"); | |
208 | ||
209 | tbl = (struct ip_vs_dh_bucket *)svc->sched_data; | |
210 | dest = ip_vs_dh_get(tbl, iph->daddr); | |
211 | if (!dest | |
212 | || !(dest->flags & IP_VS_DEST_F_AVAILABLE) | |
213 | || atomic_read(&dest->weight) <= 0 | |
214 | || is_overloaded(dest)) { | |
215 | return NULL; | |
216 | } | |
217 | ||
218 | IP_VS_DBG(6, "DH: destination IP address %u.%u.%u.%u " | |
219 | "--> server %u.%u.%u.%u:%d\n", | |
220 | NIPQUAD(iph->daddr), | |
221 | NIPQUAD(dest->addr), | |
222 | ntohs(dest->port)); | |
223 | ||
224 | return dest; | |
225 | } | |
226 | ||
227 | ||
228 | /* | |
229 | * IPVS DH Scheduler structure | |
230 | */ | |
231 | static struct ip_vs_scheduler ip_vs_dh_scheduler = | |
232 | { | |
233 | .name = "dh", | |
234 | .refcnt = ATOMIC_INIT(0), | |
235 | .module = THIS_MODULE, | |
236 | .init_service = ip_vs_dh_init_svc, | |
237 | .done_service = ip_vs_dh_done_svc, | |
238 | .update_service = ip_vs_dh_update_svc, | |
239 | .schedule = ip_vs_dh_schedule, | |
240 | }; | |
241 | ||
242 | ||
243 | static int __init ip_vs_dh_init(void) | |
244 | { | |
245 | INIT_LIST_HEAD(&ip_vs_dh_scheduler.n_list); | |
246 | return register_ip_vs_scheduler(&ip_vs_dh_scheduler); | |
247 | } | |
248 | ||
249 | ||
250 | static void __exit ip_vs_dh_cleanup(void) | |
251 | { | |
252 | unregister_ip_vs_scheduler(&ip_vs_dh_scheduler); | |
253 | } | |
254 | ||
255 | ||
256 | module_init(ip_vs_dh_init); | |
257 | module_exit(ip_vs_dh_cleanup); | |
258 | MODULE_LICENSE("GPL"); |